Hardened repository corruption handling

To improve support for the largest codebases, Sourcegraph is now significantly more resilient to repository corruption events.

We have made coordinated improvements across optimization, corruption detection, and recovery so repositories stay available and searchable in more scenarios.

Repository corruption can occur when Git objects are missing or corrupted. This can happen for many reasons, including incomplete or interrupted clones/fetches, filesystem corruption, or other disk issues. When Git objects are missing or corrupted, Sourcegraph operations that require those objects (such as listing a repository directory tree, retrieving files, resolving references, or searching commits) can fail.

Here's what changed:

1. Sourcegraph now more reliably detects corruption that occurs while repositories are being updated (e.g. git fetch).
2. Repository optimization and long-running fetch operations are now better tuned to reduce the risk of corruption during concurrent maintenance work.
3. Sourcegraph now performs a zero-downtime recovery flow (in most cases) by atomically recloning in the background, optimizing it, and atomically swapping it into place. Previously, recovery immediately deleted the on-disk repository, which disrupted repository operations and availability.

You can monitor this recovery process from the repository settings page, which now reports reclone progress directly in the UI.